Sample training based wildfire segmentation by 2D histogram θ-division with minimum error.

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Wildfire segmentation is crucial for monitoring and response.
• Existing θ-division methods based on 2D histograms and minimum error lack prior knowledge integration.
• Accurate segmentation of fire pixels in images remains a challenge.

Purpose of the Study:

• To propose a novel wildfire segmentation algorithm.
• To integrate prior knowledge into θ-division methods for improved accuracy.
• To evaluate the algorithm's efficiency on real-world wildfire images.

Main Methods:

• Developed a sample training-based 2D histogram θ-division algorithm.
• Defined a probability function of error division to evaluate segmentations.
• Determined the optimal angle θ through sample training.
• Compared performance across different color channels and selected the most suitable one.
• Combined θ-division with other methods like Gaussian Mixture Models (GMM) for enhanced accuracy.

Main Results:

• The proposed algorithm effectively segments wildfires using sample-trained θ-division and minimum error.
• Optimal angle θ was determined through empirical sample training.
• Performance analysis identified the most effective color channel for segmentation.
• A combined approach integrating θ-division with GMM further improved segmentation accuracy.
• Experimental results on real images demonstrate the algorithm's high efficiency.

Conclusions:

• The novel wildfire segmentation algorithm demonstrates significant efficiency.
• Sample training and prior knowledge integration enhance θ-division segmentation.
• The combined approach offers a robust solution for wildfire detection in images.